Related Experiment Video
Updated: Mar 12, 2026
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
43.8K
Breast parenchymal patterns in processed versus raw digital mammograms: A large population study toward assessing
Aimilia Gastounioti1, Andrew Oustimov1, Brad M Keller1
1Department of Radiology, Perelman School of Medicine, University of Pennsylvania, 3700 Hamilton Walk, Rm D702 Richards Bldg., Philadelphia, Pennsylvania 19104.
Medical Physics
|November 4, 2016
Summary
Quantitative measures of breast density and texture differ between raw and processed digital mammograms (DMs). Structural texture features show more consistent results across different DM settings, offering a robust option for breast cancer risk assessment studies.
Area of Science:
- Radiology and Medical Imaging
- Biomedical Engineering
- Oncology
Background:
- Raw digital mammograms (DMs) preserve x-ray attenuation information crucial for breast cancer risk assessment.
- Processed DMs are more commonly available but may alter quantitative imaging measures.
- Understanding differences between raw and processed DMs is vital for consistent breast cancer risk evaluation.
Purpose of the Study:
- To investigate quantitative differences in breast density and parenchymal texture measures between raw and processed DMs.
- To assess the impact of vendor and acquisition factors on these quantitative measures.
- To compare the association of these measures with breast cancer risk factors and breast tissue symmetry.
Main Methods:
- Analysis of 8458 bilateral DM pairs from 4278 women across two vendors.
- Automated software used to measure breast dense tissue area, percent density (PD), and various texture features (statistical, co-occurrence, run-length, structural).
- Statistical comparisons included Wilcoxon signed-ranks test, correlation, and linear-mixed-effects (LME) models, assessing interactions with patient and system factors.
Main Results:
- High correlations (r ≥ 0.6) observed for most density and texture features between raw and processed DMs, but significant differences (p < 0.05) exist.
- Measurement variability influenced by feature type, vendor, and acquisition settings (age, BMI, mAs/kVp).
- Structural texture features demonstrated the strongest correlations and minimal LME model interactions; measures showed weak association with Gail risk but moderate with PD, and strong bilateral symmetry.
Conclusions:
- Differences between raw and processed DM measures are feature-, vendor-, and setting-dependent.
- Structural texture features are more robust across varying DM settings.
- Findings provide a reference for future large-scale studies on mammographic features and breast cancer risk assessment using diverse DM representations.
